RuntimeError: weight tensor should be defined either for all 4 classes or no classes but got weight tensor of shape: [3] about doc2graph HOT 4 OPEN

I have the same problem as you, have you fixed this error yet?

from doc2graph.

I have the same problem as you, have you fixed this error yet?

Sorry, I couldn't find a solution. And you?

from doc2graph.

I have the same problem as you, have you fixed this error yet?

Hi， I fixed this bug by changing some code.

First: in utils compute_crossentropy_loss, add

    new_labels = labels.cpu().numpy().tolist()

    if len(set(new_labels)) == 3:
        for i in range(4):
            if i not in new_labels:
                w = w.tolist()
                w.insert(i, len(new_labels))
                break

    if len(set(new_labels)) == 1:
        for i in range(2):
            if i not in new_labels:
                w = w.tolist()
                w.insert(i, len(new_labels))
                break

Second: in utils compute_auc_mc function, changing a line

# from
labels = F.one_hot(labels).cpu().numpy()

# to 
labels = F.one_hot(labels, num_classes=2).cpu().numpy()

I hope I can help you

from doc2graph.

I have the same problem as you, have you fixed this error yet?

Hi， I fixed this bug by changing some code.

First: in utils compute_crossentropy_loss, add

    new_labels = labels.cpu().numpy().tolist()

    if len(set(new_labels)) == 3:
        for i in range(4):
            if i not in new_labels:
                w = w.tolist()
                w.insert(i, len(new_labels))
                break

    if len(set(new_labels)) == 1:
        for i in range(2):
            if i not in new_labels:
                w = w.tolist()
                w.insert(i, len(new_labels))
                break

Second: in utils compute_auc_mc function, changing a line

# from
labels = F.one_hot(labels).cpu().numpy()

# to 
labels = F.one_hot(labels, num_classes=2).cpu().numpy()

I hope I can help you

Thank you so much for your help! I really appreciate it.

from doc2graph.